One of the biggest unresolved questions about the January 6th mob assault on the U.S. Capitol last year is how much social media factored into the violence that day. One year on, just over 700 people who answered Donald Trump’s call to march on Congress are now facing charges in connection with the assault. Many of those indicted were identified through social media posts to Facebook, Twitter, and Parler, the far-right rival to mainstream platforms.
Facebook and Twitter have contended in the wake of the Capitol siege that their approach to content moderation is much more effective than so-called alternative tech, or alt-tech, platforms like Parler that brand themselves as champions of unmoderated free speech. But absent hard evidence and transparency about platform design features and about how content moderation decisions are made and implemented, it is hard to test those claims.
What we do know, one year on from January 6th, is that understanding how platform governance shapes the feedback loop between online social media and offline violence is crucial as we head toward the 2022 midterm elections. The urgency will only intensify if Trump runs for president again or otherwise involves himself in the 2024 election and succeeds in standing up his own social media company.
Recent polls indicate Americans are sharply divided about what happened on Jan. 6 and what it means for the future. But there is little question that the storming of the Capitol not only created a crisis of confidence in American democracy, but also shredded public faith in the tech industry. The attack on Congress and explosion of violent online content prompted Twitter and Facebook to ban Trump from their platforms, while Apple, Google, and Amazon all moved to cut off access to Parler.
Yet, there is a lot we still don’t know about how social media factored into the calculations and decisions of the thousands of Americans who turned up in Washington on Jan. 6 to support Trump. How did niche social media sites geared toward far-right audiences, like Parler, contribute to polarization around the 2020 presidential election and factor into the January 6th attack? What trends can be extrapolated from the Parler data, and how might this data be used to detect early warning signs of political violence? Was Parler really more culpable, in important respects, than Facebook or Twitter in fanning the flames of the #StoptheSteal movement?
Millions of Users Sign Up
Founded in 2018, Parler took off during the 2020 presidential race. In the summer and fall of that year, high-profile conservatives, including representatives of the Trump campaign, urged their supporters to join them on Parler, which billed itself as a haven for free speech with a laissez-faire approach to moderation compared to mainstream platforms.
Millions of users signed up, driving a surge of activity leading up to and following the November election. In this environment, “Stop the Steal” — the false narrative that Democrats had rigged the election in favor of Joe Biden — spread like wildfire. Election conspiracy theories drove conversations on Parler to such an extent that posts mentioning “Stop the Steal” and overall Parler activity both peaked the week after Biden was projected to win the presidency, according an analysis of Parler data we conducted using the Social Media Analysis Toolkit.
In the immediate aftermath of the attack, Facebook chief operating officer Sheryl Sandberg tried to shift the blame for the violence onto Parler. The only way to test those claims, however, is to look at the data across platforms, and there is no shortage of that. On Jan. 11, 2021, an anonymous hacktivist who uses the handle @donk_enby on Twitter, released a massive tranche of Parler data that included 1,030,523 video posts and metadata from posts made by 13 million Parler users.
Working with our colleagues at Arizona State University and New America’s Future Frontlines program, we tapped into that data and other sources to assemble the puzzle pieces of what happened online and offline leading up to the January 6 siege. In addition to the @donk_enby video and image posts and metadata, another data source consisted of 183 million Parler posts collected by an international consortium of researchers, including several from New York University. Another data stream included charging documents for more than 630 individuals whom the FBI arrested and charged with crimes related to the Capitol breach, which traced arrestees to their state and town of residence. Yet another data set consisted of location information about protests that took place across the United States in the year leading up to the January 6th attack.
Nearly 60,000 of the 1 million-plus videos @donk_enby released contained latitude and longitude coordinates in the metadata, allowing us to map those videos to specific upload locations. The greatest density of video uploads was in the Washington, DC, region, with more than 1,000 in that area alone. The second most common location was St. Louis, Missouri, with over 500 video uploads.
Geographic Breadth
However, more than any particular location, what stands out about Parler’s video content is its sheer geographic breadth. Parler users uploaded videos from many distinct locations across all 50 U.S. states, far too many to be the work of a centralized team. Many of the geotagged videos appear to be clustered around the suburbs and exurbs of major metropolitan areas. Rural areas are also well represented, particularly in more densely populated and privately owned rural areas east of the Rocky Mountains. By contrast, relatively few videos were uploaded near city centers.
Geographic diversity also increased over time. Parler users uploaded few videos to the platform before spring 2020, and until then, videos were sparsely distributed and varied in location from one month to the next. Only after the presidential campaign heated up in the summer of 2020, and video content spiked on Parler, does the data begin to show a trend toward consistent, diverse upload locations across the United States.
All this suggests Parler’s successful penetration of grassroots political life in the second half of 2020. By November, the platform and “Stop the Steal” messaging, carried on the wings of politicians and influencers, had reached thousands of communities across the United States.
The list of Americans arrested and charged with crimes related to the January 6th Capitol attack confirms that “Stop the Steal” was a nationwide phenomenon. We mined charging documents and local news reports to find states and towns of residence for 632 individuals whom the FBI had arrested as of October 2021. The resulting dataset includes arrestees from 45 states and the District of Columbia. The top three states of origin were Florida (70), Texas (57), and Pennsylvania (56).
We wondered whether proximity to social unrest over the course of 2020 may have played some role in activating those who stormed the Capitol. So, we asked Princeton University’s Bridging Divides Initiative to analyze arrestee residences against the U.S. Crisis Monitor, which records demonstrations and instances of political violence across the political spectrum.
The resulting analysis makes clear that the “Stop the Steal” movement gave geographic reach to “Stop the Steal” and suggests patterns of activation, such as proximity to protests. We found that the residences of those charged in the January 6th attack tended to cluster near areas where all types of demonstrations and counter protests took place, to a greater degree than anticipated under complete spatial randomness. In that respect, offline organized activity contributed to online activity, creating a potential feedback loop.
Patterns of Mobilization
We also learned that patterns of mobilization to violence on the platform were geographically distinct and evident well before the riot at the Capitol. Moreover, this relationship was strongest on the local rather than state level. That suggests that the proximity of protest activity prior to Jan. 6 may have been an influencing factor for a substantial number of those indicted in connection with the Capitol breach.
Understanding how protest activity near their homes may have influenced the Capitol rioters will require more research. Still, our work suggests that protest monitoring, combined with geotagged and anonymized social media content like that found in Parler’s videos, may prove useful tools for detecting early warning signs of political violence from publicly available data. In other words, these types of findings may contribute to predictive models down the line.
Still, although the sheer volume of publicly available data from Parler is impressive and holds tremendous potential for modeling, it is important to note it does have limitations. It is impossible to collect complete data from most social media platforms, and the deplatforming of Parler means that the millions of posts we and other researchers have collected represent only a snapshot in time, and likely a highly pixelated one at that.
So-called “alt-tech” platforms in particular will be important to monitor and analyze as the 2022 midterm elections and 2024 presidential elections draw near. Our research, in fact, shows Parler’s business model, platform design features, and content management system – which gave outsized influence to high-profile Republican politicians like California Congressman Devin Nunes and pro-Trump personalities like Lin Wood — combined to make the social media platform and its users especially vulnerable to high levels of manipulation and exposure to what Facebook and researchers refer to as coordinated inauthentic behavior.
Intimidating Content at a Large Scale
Coordinated inauthentic behavior may take any number of forms. In some cases, it might mean the use of bots to amplify content at a rapid rate, which can make memes and catchy hashtags like #StoptheSteal go viral. In other cases, as we saw with Russian interference in the 2016 presidential election, fake accounts can be used en masse to create a sense of ubiquity. It can also mean “brigading” — that is, cooperating across networked social media to post intimidating content at such a large scale that it drowns out the voices of intended targets. Bottom line: what makes online behavior inauthentic is when posts are made at a rate and scale that would defy the laws of physics and human biology for a singular person.
Sometimes inauthentic behavior is deployed for deceptive reasons, as we saw with Russian interference in the 2016 and 2018 elections. Sometimes it is used to help spread important information faster and farther, as we’ve seen with important announcements about COVID precautions. The problem is that, when it is coordinated on a large scale, the overall impact is to create the illusion of virality. Or it can result in the misperception that content is trending and that sentiments are being widely validated in trusted networks of users and influencers. In both cases, it plays on cognitive biases that make us vulnerable to trickery and deception by machines.
Having a rough definition of inauthentic coordinated behavior and a sense of what it looks like on a platform like Parler isn’t quite enough, however. The only way to know with any certainty about what the future impact of social media might be on America’s highly polarized politics is to examine how Parler users behaved before the platform was shut down and how the platform operated in a universe with giant supernovas like Twitter and Facebook shaping the very center of gravity on the Internet. It’s also important to understand what, if any, connections exist between online and offline behavior.
What we’ve learned so far does not bode well for the future as we turn to the 2022 midterm elections. Given that Trump is launching his own social media platform, Truth Social, and that he may run for president again in 2024 or else try to shape the outcome, there are real reasons to worry about the proliferation of far-right alt-tech applications.
It has become commonplace in academia to call social media sites like Parler an “echo chamber.” But this term fails to capture the breadth and depth of the potential impact that lightly moderated platforms like Parler can have on political discourse and extremist violence. An echo chamber evokes an image of a bunch of words and memes bouncing around wildly inside a closed container. A more apt comparison for Parler and similar alt-tech sites is a hothouse: a place where conversations run wild, become radical, and eventually spill into the rest of the Internet and American political life.
The question for Congress, the White House, and the American public now is whether the United States can afford to let thousands or millions more poisonous hothouses bloom. If we want to prevent another January 6th, policymakers, researchers, and journalists likely need to keep a keen eye turned to Parler and the unwalled gardens where the alt-tech movement has taken root.